Balancing act: the complex role of artificial intelligence in addressing burnout and healthcare workforce dynamics

Digital scribes

Digital scribes are rapidly evolving as a potential solution to reduce the burden of clinical documentation. Digital scribe is an AI technology that integrates speech recognition with NLP to synthesise provider–patient interaction, summarise it in the EHR,13 populate diagnostic fields and create billing codes. There has been an increase in private healthcare-informational technology companies that are developing and perfecting this technology.14 Early studies have shown the promise of digital scribes in improving documentation efficiency by almost 2.7-fold,15 but the clinical validity and usability of this AI-based technology is still in its infancy.

This automation not only streamlines the record-keeping process but also enhances data accuracy and accessibility. In a retrospective analysis of documented notes, Steinkamp et al showed that 50.1% of total text in patients’ notes was copied forward from prior notes.16 The redundancy in clinical documentation inundates clinicians with copious amounts of duplicated data, leading to wasted time in chart review and contributing to burnout.

Further comprehensive research is required to evaluate the impact of AI-based documentation technologies on clinician burnout reduction over the long term. For example, Nguyen et al conducted a study where digital scribes were implemented in a cancer centre. The findings indicated that the adoption of digital scribes was deemed marginally acceptable and appropriate, as well as marginally usable by oncologists. However, the study did not demonstrate a reduction in clinician burnout. Moreover, the study was constrained by a small participant pool, and it did not investigate the long-term effects of digital scribes on clinician burnout.17 As healthcare institutions grapple with increasing workplace efficiency, the impact of these technologies on reducing HCWs burnout must be a central focus. Additionally, increased workplace efficiency may have a perverse and unintentionally effect on morale as health systems push HCWs to see more patients to optimise profit margins. Developing a regulatory framework that monitors efficiency measures with HCWs well-being and patients’ outcomes will be necessary to mitigate this potential downstream effect.

Inbox management and digital health communication

Another facet of the administrative burden is the time clinicians spend responding to inbox messages and other digital forms of healthcare-related communication. The electronic inbox has become an overwhelming aspect of medical practice. Historically, managing administrative tasks was an implicit part of the job for physicians, interacting with regulators, payors and vendors. However, the shift to a digital platform has led to an unchecked increase in volume. Several factors contribute to this surge:

1. Federal regulations, such as the mandate from the Department of Health and Human Services for immediate release of test results, have doubled patient inquiries via patient portals, often about sensitive information.

2. The financial structures of healthcare, including the preference for copays during in-person visits over patient portal interactions, have encouraged more portal use for patients. The bureaucracy of prior authorisations and equipment renewals adds to the workload without additional pay.

3. Healthcare delivery organisations have shifted tasks previously handled by support staff to physicians, under the guise of ‘free’ physician time. This results in physicians burdened by tasks inconsistent with their qualification level, such as clerical work and minor clinical queries.

4. Commercial pharmacies, through automated systems designed to reduce their costs, flood inboxes with unnecessary prescription renewal requests.

Prepandemic data showed that family physicians spend 1.5 hours daily on inbox management.12 US clinicians receive almost triple the inbox messages compared with their international counterparts, with over a third being system-generated and of low value. Patient messages, while constituting only 3% of the total, demand more time and careful handling. These have increased by 157% since the pandemic’s onset and have remained high. Patients now expect immediate, real-time access to their physicians for non-urgent matters, a service that some institutions promote as a competitive advantage. Yet, the healthcare system lacks the supportive care teams and compensation models to effectively provide this service. Additionally, with increasing patient portal registrations, patients are messaging clinicians more often, thus increasing the time clinicians spend in EHRs dramatically.18

Services such as Generative AI and Co-Pilot offer promising avenues to address the issues of inbox management and digital healthcare communication. Autocompletion of text while typing or templating/drafting responses to the deluge of messages has the potential to accelerate response times and inbox clearance.19 These advanced tools leverage NLP and ML algorithms to categorise, prioritise and respond to a high volume of patient inquiries, appointment requests and other routine communications. By automating repetitive tasks, healthcare professionals can focus more on patient care, reducing response times and improving the overall efficiency of healthcare services. Furthermore, these AI systems can be trained to recognise urgent requests, ensuring critical health issues are addressed promptly, thereby enhancing patient safety and satisfaction. While the research on this subject is still emerging, a study by Ayers et al demonstrated that AI chatbot responses to patient questions are a possible solution. Ayers et al performed a cross-sectional study with 195 randomly selected patient questions from a social media forum and showed that chatbots can generate higher quality and more empathetic responses to patient questions than physicians could.20

Automated billing and coding

Implementing AI in medical billing and coding could substantially lighten the administrative burden on doctors. Medical professionals currently dedicate a significant amount of their time to billing tasks, which takes away from patient-focused activities.21 This process is multifaceted, encompassing coding, insurance claims and addressing denials, all within a tightly regulated industry. Keeping pace with regulatory changes demands ongoing education and adaptability, adding to the stress of the job. Accurate and prompt billing is critical to a practice’s revenue, and mistakes can result in significant financial repercussions. Interactions with insurers can be complex and contentious, often leading to extra work and tension. Despite the intention behind EHRs and billing systems to make the process smoother, they can at times be unwieldy and error prone.

By ensuring accuracy of billing codes and congruency between physician and nursing notes, AI can reduce chart errors, improving billing revenue cycle and the time physicians spend correcting charts. In a retrospective cohort study, Kim et al showed that NLP in combination with ML predictive modelling using Random Forest, had an 87% accuracy with the CPT billing codes generated by their billing department for elective spine surgeries.22 Emerging AI technologies like Iodine’s AwareCDI Suite have shown that they can help hospitals with better utilisation management, optimise coding and improve revenue.23 Though we are in the infancy of automation for billing practices, a potential exists for large-scale administrative savings lies ahead and health systems must balance the cost of buying AI venture-based medical coding software with the potential long-term administrative cost savings. As healthcare institutions grapple with workplace efficiency, they must invest in research and prioritisation of AI technologies that aid in reducing clinicians’ documentation, medical coding burdens and bring the HCWs back-to-the bedside.

Reducing cognitive load through information synthesis

AI excels in synthesising vast amounts of healthcare data into coherent summaries and intuitive visualisations, enhancing comprehension and decision-making for healthcare providers. It can aggregate and interpret complex medical information from various sources, presenting it in a simplified format that highlights key insights. While preliminary studies show promise, research by Ando et al brings to light the current gaps—particularly, AI’s struggle to emulate the nuanced, tacit knowledge that physicians possess, often leading to summaries that may lack the depth of human-infused clinical reasoning.25 Though the authors note that AI output editing by a clinician can bridge this gap, discharge summaries independent of this factor would be far more advantageous.

Private tech companies such as DigitalOwl, Wisedocs offer AI-digital solutions to structure medical data and provide summaries with relevant variables such as patient diagnoses, treatments, medications among other significant findings, but large-scale adaptation of these technologies with proven results has yet to be a reality. Once realised, these summaries of medical records can be traded through directed exchange or query-based exchange across health systems.

Medication pop-up alerts for allergies or drug–drug interactions have inundated clinicians with copious, with at times, irrelevant information. Liu et al studied that AI techniques could be used to have more relevant medication alerts which can reduce alert volumes by 54%,26 thus reducing the cognitive burden on clinicians. Remote monitoring devices is another example where clinicians must sift through copious information for relevant data points. AI can help swift through this data to identify relevant data points or changes in patients’ clinical status.7 If these AI-based solutions are realised, it would reduce physician’s burden to integrate various data systems, filter through unstructured medical data, reduce time spent writing discharge summaries and decrease pop-up alert fatigue, thus reducing cognitive burden and bring HCWs back to the bedside. More investment to develop and perfect this technology can result in untold gains.

Reduced cognitive burden through AI-enhanced diagnostics

AI’s role in diagnostic support also has the potential to reduce cognitive load on HCW in the realm of diagnosis and problem solving. Numerous studies highlight reduced cognitive burden through AI-enhanced diagnostics. In dermatology, AI-based technologies have diagnosed skin pathologies at nearly the same level as dermatologists for some conditions.27 Early studies in pathology indicate that a synergistic approach with AI-based technologies and human intelligence results in a far more efficient and accurate diagnostics.28 By leveraging vast datasets to analyse images, lab results and patient histories, AI algorithms can identify patterns and anomalies that may elude human detection. This support not only accelerates the diagnostic process but also elevates diagnostic accuracy, potentially leading to early detection and treatment of conditions. When health systems integrate these AI technologies into clinical workflows, there potential for increased workplace efficiency, clinical accuracy and overall reduced cognitive burden on HCWs and reduction in the stress associated with medical error.

Forecasting

Predictive analytics powered by AI can forecast patient deteriorations and improve hospital operations by being proactive instead of reactive. Winslow et al’s study on ML model for early warning for clinical deterioration has shown a decrease in-hospital mortality as it resulted in early recognition and transfer to intensive care units of sick patients.29 For physicians, the ability to anticipate and mitigate health issues before they escalate can reduce the stress and unpredictability associated with emergency interventions. This prediction goes beyond clinician burnout. In addition, AI has shown promise in anticipating patient utilisation of healthcare resources. Scheduling, registration, transporters are just a few additional examples where workflow optimisation can be found through intelligent scheduling of outpatient appointments, predicting no-shows, allocating transporters in the hospital, enhancing bed turnover and anticipating emergent use of healthcare resources.

Deskilling and over-reliance

A crucial issue at the intersection of healthcare and technology is the potential deskilling and overreliance on AI technologies, such as Large Language Models. As clinicians increasingly depend on AI for diagnostics and treatment recommendations, there is a tangible risk that their clinical skills may deteriorate.30 This dependency could impair physicians’ abilities to make independent, critical decisions, especially in situations where AI systems are unavailable, malfunction or yield erroneous results. Moreover, the dynamic nature of healthcare employment, where professionals often transition between diverse clinical settings, exacerbates this risk.31 This variability across workplaces underscores the urgency of sustaining and enhancing clinical skills in tandem with AI utilisation. The introduction of AI could also hinder the acquisition of essential skills by medical trainees. Overreliance on AI for decision-making might prevent trainees from developing the critical thinking and problem-solving abilities necessary for independent practice.32 This could result in increased anxiety and cognitive load if AI systems are not available, particularly in emergency situations.

Changing cognitive demands

The integration of AI in healthcare not only alters the types of tasks performed by physicians but also shifts the cognitive demands placed on them.33 While AI is adept at handling routine tasks efficiently, it often leaves physicians to manage more complex and challenging cases. While this shift has the potential to offer intellectual stimulation by engaging healthcare providers with high-level medical problems, the continual exposure to such high-stress scenarios without the balance of less demanding tasks could inadvertently exacerbate burnout.34 The constant demand to tackle only complex cases might also diminish job satisfaction and long-term engagement, as the variation in task difficulty—often linked to career longevity—is reduced. In addition, issues around the explainability of AI systems can exacerbate these tendencies, hindering HCWs ability to communicate effectively with patients and make informed decisions.35

Exacerbation of feelings of patient-healthcare worker alienation

Widespread integration of AI in healthcare could intensify feelings of alienation between patients and healthcare workers. As AI assumes greater responsibilities, the essential human touch in patient care might become less prevalent, potentially diminishing patient satisfaction and trust.36 Additionally, healthcare workers may feel marginalised by AI technology, concerned about job displacement or the devaluation of their professional skills. This concern is underscored by developments such as generative AI nurses, which operate at significantly lower costs compared with human nurses.37 Such cost efficiencies may pressure healthcare systems to view human workers as replaceable elements within AI-driven workflows. This is just one of many concerns that the deployment of AI could lead healthcare providers and institutions to prioritise financial efficiencies at the expense of meaningful improvements in patient care (ie, the pressure to streamline operations, reduce costs, and boost revenues might overshadow the potential of these technologies to enhance care quality or improve patient outcomes).

Equity concerns in AI integration into healthcare

The integration of AI in healthcare surfaces significant equity concerns, as it could perpetuate and even exacerbate existing health inequities on personal, systemic and structural levels. AI systems are typically trained on existing data, which mirrors the biases and disparities found in current healthcare practices.38 Consequently, AI algorithms might unintentionally reinforce these inequities, resulting in diagnostic and treatment recommendations that could disproportionately disadvantage marginalised populations. Moreover, the adoption and distribution of AI technologies across healthcare systems underscore additional equity issues. Advanced AI systems are more likely to be implemented in financially robust, urban academic hospitals, enhancing their healthcare services and efficiency.39 In contrast, rural healthcare systems and those serving historically marginalised communities may lack the resources to adopt these technologies, potentially widening the health service quality gap between urban and rural areas.

The commercial nature of AI development further complicates these equity issues. Often, the drive for profitability outweighs social responsibility, leading companies to prioritise marketability and cost-efficiency over the identification and correction of biases in AI models. This approach differs markedly from the motivations of independent researchers focused on ethical considerations and public health outcomes, who are more likely to address these biases.